net/sunrpc/xprtsock.c

   1 /*
   2  * linux/net/sunrpc/xprtsock.c
   3  *
   4  * Client-side transport implementation for sockets.
   5  *
   6  * TCP callback races fixes (C) 1998 Red Hat
   7  * TCP send fixes (C) 1998 Red Hat
   8  * TCP NFS related read + write fixes
   9  *  (C) 1999 Dave Airlie, University of Limerick, Ireland <airlied@linux.ie>
  10  *
  11  * Rewrite of larges part of the code in order to stabilize TCP stuff.
  12  * Fix behaviour when socket buffer is full.
  13  *  (C) 1999 Trond Myklebust <trond.myklebust@fys.uio.no>
  14  *
  15  * IP socket transport implementation, (C) 2005 Chuck Lever <cel@netapp.com>
  16  *
  17  * IPv6 support contributed by Gilles Quillard, Bull Open Source, 2005.
  18  *   <gilles.quillard@bull.net>
  19  */
  20
  21 #include <linux/types.h>
  22 #include <linux/slab.h>
  23 #include <linux/module.h>
  24 #include <linux/capability.h>
  25 #include <linux/pagemap.h>
  26 #include <linux/errno.h>
  27 #include <linux/socket.h>
  28 #include <linux/in.h>
  29 #include <linux/net.h>
  30 #include <linux/mm.h>
  31 #include <linux/udp.h>
  32 #include <linux/tcp.h>
  33 #include <linux/sunrpc/clnt.h>
  34 #include <linux/sunrpc/sched.h>
  35 #include <linux/sunrpc/svcsock.h>
  36 #include <linux/sunrpc/xprtsock.h>
  37 #include <linux/file.h>
  38 #ifdef CONFIG_NFS_V4_1
  39 #include <linux/sunrpc/bc_xprt.h>
  40 #endif
  41
  42 #include <net/sock.h>
  43 #include <net/checksum.h>
  44 #include <net/udp.h>
  45 #include <net/tcp.h>
  46
  47 #include "sunrpc.h"
  48 /*
  49  * xprtsock tunables
  50  */
  51 unsigned int xprt_udp_slot_table_entries = RPC_DEF_SLOT_TABLE;
  52 unsigned int xprt_tcp_slot_table_entries = RPC_DEF_SLOT_TABLE;
  53
  54 unsigned int xprt_min_resvport = RPC_DEF_MIN_RESVPORT;
  55 unsigned int xprt_max_resvport = RPC_DEF_MAX_RESVPORT;
  56
  57 #define XS_TCP_LINGER_TO        (15U * HZ)
  58 static unsigned int xs_tcp_fin_timeout __read_mostly = XS_TCP_LINGER_TO;
  59
  60 /*
  61  * We can register our own files under /proc/sys/sunrpc by
  62  * calling register_sysctl_table() again.  The files in that
  63  * directory become the union of all files registered there.
  64  *
  65  * We simply need to make sure that we don't collide with
  66  * someone else's file names!
  67  */
  68
  69 #ifdef RPC_DEBUG
  70
  71 static unsigned int min_slot_table_size = RPC_MIN_SLOT_TABLE;
  72 static unsigned int max_slot_table_size = RPC_MAX_SLOT_TABLE;
  73 static unsigned int xprt_min_resvport_limit = RPC_MIN_RESVPORT;
  74 static unsigned int xprt_max_resvport_limit = RPC_MAX_RESVPORT;
  75
  76 static struct ctl_table_header *sunrpc_table_header;
  77
  78 /*
  79  * FIXME: changing the UDP slot table size should also resize the UDP
  80  *        socket buffers for existing UDP transports
  81  */
  82 static ctl_table xs_tunables_table[] = {
  83         {
  84                 .procname       = "udp_slot_table_entries",
  85                 .data           = &xprt_udp_slot_table_entries,
  86                 .maxlen         = sizeof(unsigned int),
  87                 .mode           = 0644,
  88                 .proc_handler   = proc_dointvec_minmax,
  89                 .extra1         = &min_slot_table_size,
  90                 .extra2         = &max_slot_table_size
  91         },
  92         {
  93                 .procname       = "tcp_slot_table_entries",
  94                 .data           = &xprt_tcp_slot_table_entries,
  95                 .maxlen         = sizeof(unsigned int),
  96                 .mode           = 0644,
  97                 .proc_handler   = proc_dointvec_minmax,
  98                 .extra1         = &min_slot_table_size,
  99                 .extra2         = &max_slot_table_size
 100         },
 101         {
 102                 .procname       = "min_resvport",
 103                 .data           = &xprt_min_resvport,
 104                 .maxlen         = sizeof(unsigned int),
 105                 .mode           = 0644,
 106                 .proc_handler   = proc_dointvec_minmax,
 107                 .extra1         = &xprt_min_resvport_limit,
 108                 .extra2         = &xprt_max_resvport_limit
 109         },
 110         {
 111                 .procname       = "max_resvport",
 112                 .data           = &xprt_max_resvport,
 113                 .maxlen         = sizeof(unsigned int),
 114                 .mode           = 0644,
 115                 .proc_handler   = proc_dointvec_minmax,
 116                 .extra1         = &xprt_min_resvport_limit,
 117                 .extra2         = &xprt_max_resvport_limit
 118         },
 119         {
 120                 .procname       = "tcp_fin_timeout",
 121                 .data           = &xs_tcp_fin_timeout,
 122                 .maxlen         = sizeof(xs_tcp_fin_timeout),
 123                 .mode           = 0644,
 124                 .proc_handler   = proc_dointvec_jiffies,
 125         },
 126         { },
 127 };
 128
 129 static ctl_table sunrpc_table[] = {
 130         {
 131                 .procname       = "sunrpc",
 132                 .mode           = 0555,
 133                 .child          = xs_tunables_table
 134         },
 135         { },
 136 };
 137
 138 #endif
 139
 140 /*
 141  * Time out for an RPC UDP socket connect.  UDP socket connects are
 142  * synchronous, but we set a timeout anyway in case of resource
 143  * exhaustion on the local host.
 144  */
 145 #define XS_UDP_CONN_TO          (5U * HZ)
 146
 147 /*
 148  * Wait duration for an RPC TCP connection to be established.  Solaris
 149  * NFS over TCP uses 60 seconds, for example, which is in line with how
 150  * long a server takes to reboot.
 151  */
 152 #define XS_TCP_CONN_TO          (60U * HZ)
 153
 154 /*
 155  * Wait duration for a reply from the RPC portmapper.
 156  */
 157 #define XS_BIND_TO              (60U * HZ)
 158
 159 /*
 160  * Delay if a UDP socket connect error occurs.  This is most likely some
 161  * kind of resource problem on the local host.
 162  */
 163 #define XS_UDP_REEST_TO         (2U * HZ)
 164
 165 /*
 166  * The reestablish timeout allows clients to delay for a bit before attempting
 167  * to reconnect to a server that just dropped our connection.
 168  *
 169  * We implement an exponential backoff when trying to reestablish a TCP
 170  * transport connection with the server.  Some servers like to drop a TCP
 171  * connection when they are overworked, so we start with a short timeout and
 172  * increase over time if the server is down or not responding.
 173  */
 174 #define XS_TCP_INIT_REEST_TO    (3U * HZ)
 175 #define XS_TCP_MAX_REEST_TO     (5U * 60 * HZ)
 176
 177 /*
 178  * TCP idle timeout; client drops the transport socket if it is idle
 179  * for this long.  Note that we also timeout UDP sockets to prevent
 180  * holding port numbers when there is no RPC traffic.
 181  */
 182 #define XS_IDLE_DISC_TO         (5U * 60 * HZ)
 183
 184 #ifdef RPC_DEBUG
 185 # undef  RPC_DEBUG_DATA
 186 # define RPCDBG_FACILITY        RPCDBG_TRANS
 187 #endif
 188
 189 #ifdef RPC_DEBUG_DATA
 190 static void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 191 {
 192         u8 *buf = (u8 *) packet;
 193         int j;
 194
 195         dprintk("RPC:       %s\n", msg);
 196         for (j = 0; j < count && j < 128; j += 4) {
 197                 if (!(j & 31)) {
 198                         if (j)
 199                                 dprintk("\n");
 200                         dprintk("0x%04x ", j);
 201                 }
 202                 dprintk("%02x%02x%02x%02x ",
 203                         buf[j], buf[j+1], buf[j+2], buf[j+3]);
 204         }
 205         dprintk("\n");
 206 }
 207 #else
 208 static inline void xs_pktdump(char *msg, u32 *packet, unsigned int count)
 209 {
 210         /* NOP */
 211 }
 212 #endif
 213
 214 struct sock_xprt {
 215         struct rpc_xprt         xprt;
 216
 217         /*
 218          * Network layer
 219          */
 220         struct socket *         sock;
 221         struct sock *           inet;
 222
 223         /*
 224          * State of TCP reply receive
 225          */
 226         __be32                  tcp_fraghdr,
 227                                 tcp_xid;
 228
 229         u32                     tcp_offset,
 230                                 tcp_reclen;
 231
 232         unsigned long           tcp_copied,
 233                                 tcp_flags;
 234
 235         /*
 236          * Connection of transports
 237          */
 238         struct delayed_work     connect_worker;
 239         struct sockaddr_storage srcaddr;
 240         unsigned short          srcport;
 241
 242         /*
 243          * UDP socket buffer size parameters
 244          */
 245         size_t                  rcvsize,
 246                                 sndsize;
 247
 248         /*
 249          * Saved socket callback addresses
 250          */
 251         void                    (*old_data_ready)(struct sock *, int);
 252         void                    (*old_state_change)(struct sock *);
 253         void                    (*old_write_space)(struct sock *);
 254         void                    (*old_error_report)(struct sock *);
 255 };
 256
 257 /*
 258  * TCP receive state flags
 259  */
 260 #define TCP_RCV_LAST_FRAG       (1UL << 0)
 261 #define TCP_RCV_COPY_FRAGHDR    (1UL << 1)
 262 #define TCP_RCV_COPY_XID        (1UL << 2)
 263 #define TCP_RCV_COPY_DATA       (1UL << 3)
 264 #define TCP_RCV_READ_CALLDIR    (1UL << 4)
 265 #define TCP_RCV_COPY_CALLDIR    (1UL << 5)
 266
 267 /*
 268  * TCP RPC flags
 269  */
 270 #define TCP_RPC_REPLY           (1UL << 6)
 271
 272 static inline struct sockaddr *xs_addr(struct rpc_xprt *xprt)
 273 {
 274         return (struct sockaddr *) &xprt->addr;
 275 }
 276
 277 static inline struct sockaddr_in *xs_addr_in(struct rpc_xprt *xprt)
 278 {
 279         return (struct sockaddr_in *) &xprt->addr;
 280 }
 281
 282 static inline struct sockaddr_in6 *xs_addr_in6(struct rpc_xprt *xprt)
 283 {
 284         return (struct sockaddr_in6 *) &xprt->addr;
 285 }
 286
 287 static void xs_format_common_peer_addresses(struct rpc_xprt *xprt)
 288 {
 289         struct sockaddr *sap = xs_addr(xprt);
 290         struct sockaddr_in6 *sin6;
 291         struct sockaddr_in *sin;
 292         char buf[128];
 293
 294         (void)rpc_ntop(sap, buf, sizeof(buf));
 295         xprt->address_strings[RPC_DISPLAY_ADDR] = kstrdup(buf, GFP_KERNEL);
 296
 297         switch (sap->sa_family) {
 298         case AF_INET:
 299                 sin = xs_addr_in(xprt);
 300                 snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
 301                 break;
 302         case AF_INET6:
 303                 sin6 = xs_addr_in6(xprt);
 304                 snprintf(buf, sizeof(buf), "%pi6", &sin6->sin6_addr);
 305                 break;
 306         default:
 307                 BUG();
 308         }
 309         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
 310 }
 311
 312 static void xs_format_common_peer_ports(struct rpc_xprt *xprt)
 313 {
 314         struct sockaddr *sap = xs_addr(xprt);
 315         char buf[128];
 316
 317         snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
 318         xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
 319
 320         snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
 321         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
 322 }
 323
 324 static void xs_format_peer_addresses(struct rpc_xprt *xprt,
 325                                      const char *protocol,
 326                                      const char *netid)
 327 {
 328         xprt->address_strings[RPC_DISPLAY_PROTO] = protocol;
 329         xprt->address_strings[RPC_DISPLAY_NETID] = netid;
 330         xs_format_common_peer_addresses(xprt);
 331         xs_format_common_peer_ports(xprt);
 332 }
 333
 334 static void xs_update_peer_port(struct rpc_xprt *xprt)
 335 {
 336         kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
 337         kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
 338
 339         xs_format_common_peer_ports(xprt);
 340 }
 341
 342 static void xs_free_peer_addresses(struct rpc_xprt *xprt)
 343 {
 344         unsigned int i;
 345
 346         for (i = 0; i < RPC_DISPLAY_MAX; i++)
 347                 switch (i) {
 348                 case RPC_DISPLAY_PROTO:
 349                 case RPC_DISPLAY_NETID:
 350                         continue;
 351                 default:
 352                         kfree(xprt->address_strings[i]);
 353                 }
 354 }
 355
 356 #define XS_SENDMSG_FLAGS        (MSG_DONTWAIT | MSG_NOSIGNAL)
 357
 358 static int xs_send_kvec(struct socket *sock, struct sockaddr *addr, int addrlen, struct kvec *vec, unsigned int base, int more)
 359 {
 360         struct msghdr msg = {
 361                 .msg_name       = addr,
 362                 .msg_namelen    = addrlen,
 363                 .msg_flags      = XS_SENDMSG_FLAGS | (more ? MSG_MORE : 0),
 364         };
 365         struct kvec iov = {
 366                 .iov_base       = vec->iov_base + base,
 367                 .iov_len        = vec->iov_len - base,
 368         };
 369
 370         if (iov.iov_len != 0)
 371                 return kernel_sendmsg(sock, &msg, &iov, 1, iov.iov_len);
 372         return kernel_sendmsg(sock, &msg, NULL, 0, 0);
 373 }
 374
 375 static int xs_send_pagedata(struct socket *sock, struct xdr_buf *xdr, unsigned int base, int more)
 376 {
 377         struct page **ppage;
 378         unsigned int remainder;
 379         int err, sent = 0;
 380
 381         remainder = xdr->page_len - base;
 382         base += xdr->page_base;
 383         ppage = xdr->pages + (base >> PAGE_SHIFT);
 384         base &= ~PAGE_MASK;
 385         for(;;) {
 386                 unsigned int len = min_t(unsigned int, PAGE_SIZE - base, remainder);
 387                 int flags = XS_SENDMSG_FLAGS;
 388
 389                 remainder -= len;
 390                 if (remainder != 0 || more)
 391                         flags |= MSG_MORE;
 392                 err = sock->ops->sendpage(sock, *ppage, base, len, flags);
 393                 if (remainder == 0 || err != len)
 394                         break;
 395                 sent += err;
 396                 ppage++;
 397                 base = 0;
 398         }
 399         if (sent == 0)
 400                 return err;
 401         if (err > 0)
 402                 sent += err;
 403         return sent;
 404 }
 405
 406 /**
 407  * xs_sendpages - write pages directly to a socket
 408  * @sock: socket to send on
 409  * @addr: UDP only -- address of destination
 410  * @addrlen: UDP only -- length of destination address
 411  * @xdr: buffer containing this request
 412  * @base: starting position in the buffer
 413  *
 414  */
 415 static int xs_sendpages(struct socket *sock, struct sockaddr *addr, int addrlen, struct xdr_buf *xdr, unsigned int base)
 416 {
 417         unsigned int remainder = xdr->len - base;
 418         int err, sent = 0;
 419
 420         if (unlikely(!sock))
 421                 return -ENOTSOCK;
 422
 423         clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags);
 424         if (base != 0) {
 425                 addr = NULL;
 426                 addrlen = 0;
 427         }
 428
 429         if (base < xdr->head[0].iov_len || addr != NULL) {
 430                 unsigned int len = xdr->head[0].iov_len - base;
 431                 remainder -= len;
 432                 err = xs_send_kvec(sock, addr, addrlen, &xdr->head[0], base, remainder != 0);
 433                 if (remainder == 0 || err != len)
 434                         goto out;
 435                 sent += err;
 436                 base = 0;
 437         } else
 438                 base -= xdr->head[0].iov_len;
 439
 440         if (base < xdr->page_len) {
 441                 unsigned int len = xdr->page_len - base;
 442                 remainder -= len;
 443                 err = xs_send_pagedata(sock, xdr, base, remainder != 0);
 444                 if (remainder == 0 || err != len)
 445                         goto out;
 446                 sent += err;
 447                 base = 0;
 448         } else
 449                 base -= xdr->page_len;
 450
 451         if (base >= xdr->tail[0].iov_len)
 452                 return sent;
 453         err = xs_send_kvec(sock, NULL, 0, &xdr->tail[0], base, 0);
 454 out:
 455         if (sent == 0)
 456                 return err;
 457         if (err > 0)
 458                 sent += err;
 459         return sent;
 460 }
 461
 462 static void xs_nospace_callback(struct rpc_task *task)
 463 {
 464         struct sock_xprt *transport = container_of(task->tk_rqstp->rq_xprt, struct sock_xprt, xprt);
 465
 466         transport->inet->sk_write_pending--;
 467         clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 468 }
 469
 470 /**
 471  * xs_nospace - place task on wait queue if transmit was incomplete
 472  * @task: task to put to sleep
 473  *
 474  */
 475 static int xs_nospace(struct rpc_task *task)
 476 {
 477         struct rpc_rqst *req = task->tk_rqstp;
 478         struct rpc_xprt *xprt = req->rq_xprt;
 479         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 480         int ret = 0;
 481
 482         dprintk("RPC: %5u xmit incomplete (%u left of %u)\n",
 483                         task->tk_pid, req->rq_slen - req->rq_bytes_sent,
 484                         req->rq_slen);
 485
 486         /* Protect against races with write_space */
 487         spin_lock_bh(&xprt->transport_lock);
 488
 489         /* Don't race with disconnect */
 490         if (xprt_connected(xprt)) {
 491                 if (test_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags)) {
 492                         ret = -EAGAIN;
 493                         /*
 494                          * Notify TCP that we're limited by the application
 495                          * window size
 496                          */
 497                         set_bit(SOCK_NOSPACE, &transport->sock->flags);
 498                         transport->inet->sk_write_pending++;
 499                         /* ...and wait for more buffer space */
 500                         xprt_wait_for_buffer_space(task, xs_nospace_callback);
 501                 }
 502         } else {
 503                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 504                 ret = -ENOTCONN;
 505         }
 506
 507         spin_unlock_bh(&xprt->transport_lock);
 508         return ret;
 509 }
 510
 511 /**
 512  * xs_udp_send_request - write an RPC request to a UDP socket
 513  * @task: address of RPC task that manages the state of an RPC request
 514  *
 515  * Return values:
 516  *        0:    The request has been sent
 517  *   EAGAIN:    The socket was blocked, please call again later to
 518  *              complete the request
 519  * ENOTCONN:    Caller needs to invoke connect logic then call again
 520  *    other:    Some other error occured, the request was not sent
 521  */
 522 static int xs_udp_send_request(struct rpc_task *task)
 523 {
 524         struct rpc_rqst *req = task->tk_rqstp;
 525         struct rpc_xprt *xprt = req->rq_xprt;
 526         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 527         struct xdr_buf *xdr = &req->rq_snd_buf;
 528         int status;
 529
 530         xs_pktdump("packet data:",
 531                                 req->rq_svec->iov_base,
 532                                 req->rq_svec->iov_len);
 533
 534         if (!xprt_bound(xprt))
 535                 return -ENOTCONN;
 536         status = xs_sendpages(transport->sock,
 537                               xs_addr(xprt),
 538                               xprt->addrlen, xdr,
 539                               req->rq_bytes_sent);
 540
 541         dprintk("RPC:       xs_udp_send_request(%u) = %d\n",
 542                         xdr->len - req->rq_bytes_sent, status);
 543
 544         if (status >= 0) {
 545                 task->tk_bytes_sent += status;
 546                 if (status >= req->rq_slen)
 547                         return 0;
 548                 /* Still some bytes left; set up for a retry later. */
 549                 status = -EAGAIN;
 550         }
 551         if (!transport->sock)
 552                 goto out;
 553
 554         switch (status) {
 555         case -ENOTSOCK:
 556                 status = -ENOTCONN;
 557                 /* Should we call xs_close() here? */
 558                 break;
 559         case -EAGAIN:
 560                 status = xs_nospace(task);
 561                 break;
 562         default:
 563                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 564                         -status);
 565         case -ENETUNREACH:
 566         case -EPIPE:
 567         case -ECONNREFUSED:
 568                 /* When the server has died, an ICMP port unreachable message
 569                  * prompts ECONNREFUSED. */
 570                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 571         }
 572 out:
 573         return status;
 574 }
 575
 576 /**
 577  * xs_tcp_shutdown - gracefully shut down a TCP socket
 578  * @xprt: transport
 579  *
 580  * Initiates a graceful shutdown of the TCP socket by calling the
 581  * equivalent of shutdown(SHUT_WR);
 582  */
 583 static void xs_tcp_shutdown(struct rpc_xprt *xprt)
 584 {
 585         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 586         struct socket *sock = transport->sock;
 587
 588         if (sock != NULL)
 589                 kernel_sock_shutdown(sock, SHUT_WR);
 590 }
 591
 592 static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
 593 {
 594         u32 reclen = buf->len - sizeof(rpc_fraghdr);
 595         rpc_fraghdr *base = buf->head[0].iov_base;
 596         *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
 597 }
 598
 599 /**
 600  * xs_tcp_send_request - write an RPC request to a TCP socket
 601  * @task: address of RPC task that manages the state of an RPC request
 602  *
 603  * Return values:
 604  *        0:    The request has been sent
 605  *   EAGAIN:    The socket was blocked, please call again later to
 606  *              complete the request
 607  * ENOTCONN:    Caller needs to invoke connect logic then call again
 608  *    other:    Some other error occured, the request was not sent
 609  *
 610  * XXX: In the case of soft timeouts, should we eventually give up
 611  *      if sendmsg is not able to make progress?
 612  */
 613 static int xs_tcp_send_request(struct rpc_task *task)
 614 {
 615         struct rpc_rqst *req = task->tk_rqstp;
 616         struct rpc_xprt *xprt = req->rq_xprt;
 617         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 618         struct xdr_buf *xdr = &req->rq_snd_buf;
 619         int status;
 620
 621         xs_encode_tcp_record_marker(&req->rq_snd_buf);
 622
 623         xs_pktdump("packet data:",
 624                                 req->rq_svec->iov_base,
 625                                 req->rq_svec->iov_len);
 626
 627         /* Continue transmitting the packet/record. We must be careful
 628          * to cope with writespace callbacks arriving _after_ we have
 629          * called sendmsg(). */
 630         while (1) {
 631                 status = xs_sendpages(transport->sock,
 632                                         NULL, 0, xdr, req->rq_bytes_sent);
 633
 634                 dprintk("RPC:       xs_tcp_send_request(%u) = %d\n",
 635                                 xdr->len - req->rq_bytes_sent, status);
 636
 637                 if (unlikely(status < 0))
 638                         break;
 639
 640                 /* If we've sent the entire packet, immediately
 641                  * reset the count of bytes sent. */
 642                 req->rq_bytes_sent += status;
 643                 task->tk_bytes_sent += status;
 644                 if (likely(req->rq_bytes_sent >= req->rq_slen)) {
 645                         req->rq_bytes_sent = 0;
 646                         return 0;
 647                 }
 648
 649                 if (status != 0)
 650                         continue;
 651                 status = -EAGAIN;
 652                 break;
 653         }
 654         if (!transport->sock)
 655                 goto out;
 656
 657         switch (status) {
 658         case -ENOTSOCK:
 659                 status = -ENOTCONN;
 660                 /* Should we call xs_close() here? */
 661                 break;
 662         case -EAGAIN:
 663                 status = xs_nospace(task);
 664                 break;
 665         default:
 666                 dprintk("RPC:       sendmsg returned unrecognized error %d\n",
 667                         -status);
 668         case -ECONNRESET:
 669         case -EPIPE:
 670                 xs_tcp_shutdown(xprt);
 671         case -ECONNREFUSED:
 672         case -ENOTCONN:
 673                 clear_bit(SOCK_ASYNC_NOSPACE, &transport->sock->flags);
 674         }
 675 out:
 676         return status;
 677 }
 678
 679 /**
 680  * xs_tcp_release_xprt - clean up after a tcp transmission
 681  * @xprt: transport
 682  * @task: rpc task
 683  *
 684  * This cleans up if an error causes us to abort the transmission of a request.
 685  * In this case, the socket may need to be reset in order to avoid confusing
 686  * the server.
 687  */
 688 static void xs_tcp_release_xprt(struct rpc_xprt *xprt, struct rpc_task *task)
 689 {
 690         struct rpc_rqst *req;
 691
 692         if (task != xprt->snd_task)
 693                 return;
 694         if (task == NULL)
 695                 goto out_release;
 696         req = task->tk_rqstp;
 697         if (req->rq_bytes_sent == 0)
 698                 goto out_release;
 699         if (req->rq_bytes_sent == req->rq_snd_buf.len)
 700                 goto out_release;
 701         set_bit(XPRT_CLOSE_WAIT, &task->tk_xprt->state);
 702 out_release:
 703         xprt_release_xprt(xprt, task);
 704 }
 705
 706 static void xs_save_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 707 {
 708         transport->old_data_ready = sk->sk_data_ready;
 709         transport->old_state_change = sk->sk_state_change;
 710         transport->old_write_space = sk->sk_write_space;
 711         transport->old_error_report = sk->sk_error_report;
 712 }
 713
 714 static void xs_restore_old_callbacks(struct sock_xprt *transport, struct sock *sk)
 715 {
 716         sk->sk_data_ready = transport->old_data_ready;
 717         sk->sk_state_change = transport->old_state_change;
 718         sk->sk_write_space = transport->old_write_space;
 719         sk->sk_error_report = transport->old_error_report;
 720 }
 721
 722 static void xs_reset_transport(struct sock_xprt *transport)
 723 {
 724         struct socket *sock = transport->sock;
 725         struct sock *sk = transport->inet;
 726
 727         if (sk == NULL)
 728                 return;
 729
 730         write_lock_bh(&sk->sk_callback_lock);
 731         transport->inet = NULL;
 732         transport->sock = NULL;
 733
 734         sk->sk_user_data = NULL;
 735
 736         xs_restore_old_callbacks(transport, sk);
 737         write_unlock_bh(&sk->sk_callback_lock);
 738
 739         sk->sk_no_check = 0;
 740
 741         sock_release(sock);
 742 }
 743
 744 /**
 745  * xs_close - close a socket
 746  * @xprt: transport
 747  *
 748  * This is used when all requests are complete; ie, no DRC state remains
 749  * on the server we want to save.
 750  *
 751  * The caller _must_ be holding XPRT_LOCKED in order to avoid issues with
 752  * xs_reset_transport() zeroing the socket from underneath a writer.
 753  */
 754 static void xs_close(struct rpc_xprt *xprt)
 755 {
 756         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 757
 758         dprintk("RPC:       xs_close xprt %p\n", xprt);
 759
 760         xs_reset_transport(transport);
 761         xprt->reestablish_timeout = 0;
 762
 763         smp_mb__before_clear_bit();
 764         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
 765         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
 766         clear_bit(XPRT_CLOSING, &xprt->state);
 767         smp_mb__after_clear_bit();
 768         xprt_disconnect_done(xprt);
 769 }
 770
 771 static void xs_tcp_close(struct rpc_xprt *xprt)
 772 {
 773         if (test_and_clear_bit(XPRT_CONNECTION_CLOSE, &xprt->state))
 774                 xs_close(xprt);
 775         else
 776                 xs_tcp_shutdown(xprt);
 777 }
 778
 779 /**
 780  * xs_destroy - prepare to shutdown a transport
 781  * @xprt: doomed transport
 782  *
 783  */
 784 static void xs_destroy(struct rpc_xprt *xprt)
 785 {
 786         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 787
 788         dprintk("RPC:       xs_destroy xprt %p\n", xprt);
 789
 790         cancel_rearming_delayed_work(&transport->connect_worker);
 791
 792         xs_close(xprt);
 793         xs_free_peer_addresses(xprt);
 794         kfree(xprt->slot);
 795         kfree(xprt);
 796         module_put(THIS_MODULE);
 797 }
 798
 799 static inline struct rpc_xprt *xprt_from_sock(struct sock *sk)
 800 {
 801         return (struct rpc_xprt *) sk->sk_user_data;
 802 }
 803
 804 /**
 805  * xs_udp_data_ready - "data ready" callback for UDP sockets
 806  * @sk: socket with data to read
 807  * @len: how much data to read
 808  *
 809  */
 810 static void xs_udp_data_ready(struct sock *sk, int len)
 811 {
 812         struct rpc_task *task;
 813         struct rpc_xprt *xprt;
 814         struct rpc_rqst *rovr;
 815         struct sk_buff *skb;
 816         int err, repsize, copied;
 817         u32 _xid;
 818         __be32 *xp;
 819
 820         read_lock(&sk->sk_callback_lock);
 821         dprintk("RPC:       xs_udp_data_ready...\n");
 822         if (!(xprt = xprt_from_sock(sk)))
 823                 goto out;
 824
 825         if ((skb = skb_recv_datagram(sk, 0, 1, &err)) == NULL)
 826                 goto out;
 827
 828         if (xprt->shutdown)
 829                 goto dropit;
 830
 831         repsize = skb->len - sizeof(struct udphdr);
 832         if (repsize < 4) {
 833                 dprintk("RPC:       impossible RPC reply size %d!\n", repsize);
 834                 goto dropit;
 835         }
 836
 837         /* Copy the XID from the skb... */
 838         xp = skb_header_pointer(skb, sizeof(struct udphdr),
 839                                 sizeof(_xid), &_xid);
 840         if (xp == NULL)
 841                 goto dropit;
 842
 843         /* Look up and lock the request corresponding to the given XID */
 844         spin_lock(&xprt->transport_lock);
 845         rovr = xprt_lookup_rqst(xprt, *xp);
 846         if (!rovr)
 847                 goto out_unlock;
 848         task = rovr->rq_task;
 849
 850         if ((copied = rovr->rq_private_buf.buflen) > repsize)
 851                 copied = repsize;
 852
 853         /* Suck it into the iovec, verify checksum if not done by hw. */
 854         if (csum_partial_copy_to_xdr(&rovr->rq_private_buf, skb)) {
 855                 UDPX_INC_STATS_BH(sk, UDP_MIB_INERRORS);
 856                 goto out_unlock;
 857         }
 858
 859         UDPX_INC_STATS_BH(sk, UDP_MIB_INDATAGRAMS);
 860
 861         /* Something worked... */
 862         dst_confirm(skb_dst(skb));
 863
 864         xprt_adjust_cwnd(task, copied);
 865         xprt_update_rtt(task);
 866         xprt_complete_rqst(task, copied);
 867
 868  out_unlock:
 869         spin_unlock(&xprt->transport_lock);
 870  dropit:
 871         skb_free_datagram(sk, skb);
 872  out:
 873         read_unlock(&sk->sk_callback_lock);
 874 }
 875
 876 static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, struct xdr_skb_reader *desc)
 877 {
 878         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
 879         size_t len, used;
 880         char *p;
 881
 882         p = ((char *) &transport->tcp_fraghdr) + transport->tcp_offset;
 883         len = sizeof(transport->tcp_fraghdr) - transport->tcp_offset;
 884         used = xdr_skb_read_bits(desc, p, len);
 885         transport->tcp_offset += used;
 886         if (used != len)
 887                 return;
 888
 889         transport->tcp_reclen = ntohl(transport->tcp_fraghdr);
 890         if (transport->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
 891                 transport->tcp_flags |= TCP_RCV_LAST_FRAG;
 892         else
 893                 transport->tcp_flags &= ~TCP_RCV_LAST_FRAG;
 894         transport->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
 895
 896         transport->tcp_flags &= ~TCP_RCV_COPY_FRAGHDR;
 897         transport->tcp_offset = 0;
 898
 899         /* Sanity check of the record length */
 900         if (unlikely(transport->tcp_reclen < 8)) {
 901                 dprintk("RPC:       invalid TCP record fragment length\n");
 902                 xprt_force_disconnect(xprt);
 903                 return;
 904         }
 905         dprintk("RPC:       reading TCP record fragment of length %d\n",
 906                         transport->tcp_reclen);
 907 }
 908
 909 static void xs_tcp_check_fraghdr(struct sock_xprt *transport)
 910 {
 911         if (transport->tcp_offset == transport->tcp_reclen) {
 912                 transport->tcp_flags |= TCP_RCV_COPY_FRAGHDR;
 913                 transport->tcp_offset = 0;
 914                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG) {
 915                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
 916                         transport->tcp_flags |= TCP_RCV_COPY_XID;
 917                         transport->tcp_copied = 0;
 918                 }
 919         }
 920 }
 921
 922 static inline void xs_tcp_read_xid(struct sock_xprt *transport, struct xdr_skb_reader *desc)
 923 {
 924         size_t len, used;
 925         char *p;
 926
 927         len = sizeof(transport->tcp_xid) - transport->tcp_offset;
 928         dprintk("RPC:       reading XID (%Zu bytes)\n", len);
 929         p = ((char *) &transport->tcp_xid) + transport->tcp_offset;
 930         used = xdr_skb_read_bits(desc, p, len);
 931         transport->tcp_offset += used;
 932         if (used != len)
 933                 return;
 934         transport->tcp_flags &= ~TCP_RCV_COPY_XID;
 935         transport->tcp_flags |= TCP_RCV_READ_CALLDIR;
 936         transport->tcp_copied = 4;
 937         dprintk("RPC:       reading %s XID %08x\n",
 938                         (transport->tcp_flags & TCP_RPC_REPLY) ? "reply for"
 939                                                               : "request with",
 940                         ntohl(transport->tcp_xid));
 941         xs_tcp_check_fraghdr(transport);
 942 }
 943
 944 static inline void xs_tcp_read_calldir(struct sock_xprt *transport,
 945                                        struct xdr_skb_reader *desc)
 946 {
 947         size_t len, used;
 948         u32 offset;
 949         __be32  calldir;
 950
 951         /*
 952          * We want transport->tcp_offset to be 8 at the end of this routine
 953          * (4 bytes for the xid and 4 bytes for the call/reply flag).
 954          * When this function is called for the first time,
 955          * transport->tcp_offset is 4 (after having already read the xid).
 956          */
 957         offset = transport->tcp_offset - sizeof(transport->tcp_xid);
 958         len = sizeof(calldir) - offset;
 959         dprintk("RPC:       reading CALL/REPLY flag (%Zu bytes)\n", len);
 960         used = xdr_skb_read_bits(desc, &calldir, len);
 961         transport->tcp_offset += used;
 962         if (used != len)
 963                 return;
 964         transport->tcp_flags &= ~TCP_RCV_READ_CALLDIR;
 965         transport->tcp_flags |= TCP_RCV_COPY_CALLDIR;
 966         transport->tcp_flags |= TCP_RCV_COPY_DATA;
 967         /*
 968          * We don't yet have the XDR buffer, so we will write the calldir
 969          * out after we get the buffer from the 'struct rpc_rqst'
 970          */
 971         if (ntohl(calldir) == RPC_REPLY)
 972                 transport->tcp_flags |= TCP_RPC_REPLY;
 973         else
 974                 transport->tcp_flags &= ~TCP_RPC_REPLY;
 975         dprintk("RPC:       reading %s CALL/REPLY flag %08x\n",
 976                         (transport->tcp_flags & TCP_RPC_REPLY) ?
 977                                 "reply for" : "request with", calldir);
 978         xs_tcp_check_fraghdr(transport);
 979 }
 980
 981 static inline void xs_tcp_read_common(struct rpc_xprt *xprt,
 982                                      struct xdr_skb_reader *desc,
 983                                      struct rpc_rqst *req)
 984 {
 985         struct sock_xprt *transport =
 986                                 container_of(xprt, struct sock_xprt, xprt);
 987         struct xdr_buf *rcvbuf;
 988         size_t len;
 989         ssize_t r;
 990
 991         rcvbuf = &req->rq_private_buf;
 992
 993         if (transport->tcp_flags & TCP_RCV_COPY_CALLDIR) {
 994                 /*
 995                  * Save the RPC direction in the XDR buffer
 996                  */
 997                 __be32  calldir = transport->tcp_flags & TCP_RPC_REPLY ?
 998                                         htonl(RPC_REPLY) : 0;
 999
1000                 memcpy(rcvbuf->head[0].iov_base + transport->tcp_copied,
1001                         &calldir, sizeof(calldir));
1002                 transport->tcp_copied += sizeof(calldir);
1003                 transport->tcp_flags &= ~TCP_RCV_COPY_CALLDIR;
1004         }
1005
1006         len = desc->count;
1007         if (len > transport->tcp_reclen - transport->tcp_offset) {
1008                 struct xdr_skb_reader my_desc;
1009
1010                 len = transport->tcp_reclen - transport->tcp_offset;
1011                 memcpy(&my_desc, desc, sizeof(my_desc));
1012                 my_desc.count = len;
1013                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1014                                           &my_desc, xdr_skb_read_bits);
1015                 desc->count -= r;
1016                 desc->offset += r;
1017         } else
1018                 r = xdr_partial_copy_from_skb(rcvbuf, transport->tcp_copied,
1019                                           desc, xdr_skb_read_bits);
1020
1021         if (r > 0) {
1022                 transport->tcp_copied += r;
1023                 transport->tcp_offset += r;
1024         }
1025         if (r != len) {
1026                 /* Error when copying to the receive buffer,
1027                  * usually because we weren't able to allocate
1028                  * additional buffer pages. All we can do now
1029                  * is turn off TCP_RCV_COPY_DATA, so the request
1030                  * will not receive any additional updates,
1031                  * and time out.
1032                  * Any remaining data from this record will
1033                  * be discarded.
1034                  */
1035                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1036                 dprintk("RPC:       XID %08x truncated request\n",
1037                                 ntohl(transport->tcp_xid));
1038                 dprintk("RPC:       xprt = %p, tcp_copied = %lu, "
1039                                 "tcp_offset = %u, tcp_reclen = %u\n",
1040                                 xprt, transport->tcp_copied,
1041                                 transport->tcp_offset, transport->tcp_reclen);
1042                 return;
1043         }
1044
1045         dprintk("RPC:       XID %08x read %Zd bytes\n",
1046                         ntohl(transport->tcp_xid), r);
1047         dprintk("RPC:       xprt = %p, tcp_copied = %lu, tcp_offset = %u, "
1048                         "tcp_reclen = %u\n", xprt, transport->tcp_copied,
1049                         transport->tcp_offset, transport->tcp_reclen);
1050
1051         if (transport->tcp_copied == req->rq_private_buf.buflen)
1052                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1053         else if (transport->tcp_offset == transport->tcp_reclen) {
1054                 if (transport->tcp_flags & TCP_RCV_LAST_FRAG)
1055                         transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1056         }
1057
1058         return;
1059 }
1060
1061 /*
1062  * Finds the request corresponding to the RPC xid and invokes the common
1063  * tcp read code to read the data.
1064  */
1065 static inline int xs_tcp_read_reply(struct rpc_xprt *xprt,
1066                                     struct xdr_skb_reader *desc)
1067 {
1068         struct sock_xprt *transport =
1069                                 container_of(xprt, struct sock_xprt, xprt);
1070         struct rpc_rqst *req;
1071
1072         dprintk("RPC:       read reply XID %08x\n", ntohl(transport->tcp_xid));
1073
1074         /* Find and lock the request corresponding to this xid */
1075         spin_lock(&xprt->transport_lock);
1076         req = xprt_lookup_rqst(xprt, transport->tcp_xid);
1077         if (!req) {
1078                 dprintk("RPC:       XID %08x request not found!\n",
1079                                 ntohl(transport->tcp_xid));
1080                 spin_unlock(&xprt->transport_lock);
1081                 return -1;
1082         }
1083
1084         xs_tcp_read_common(xprt, desc, req);
1085
1086         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA))
1087                 xprt_complete_rqst(req->rq_task, transport->tcp_copied);
1088
1089         spin_unlock(&xprt->transport_lock);
1090         return 0;
1091 }
1092
1093 #if defined(CONFIG_NFS_V4_1)
1094 /*
1095  * Obtains an rpc_rqst previously allocated and invokes the common
1096  * tcp read code to read the data.  The result is placed in the callback
1097  * queue.
1098  * If we're unable to obtain the rpc_rqst we schedule the closing of the
1099  * connection and return -1.
1100  */
1101 static inline int xs_tcp_read_callback(struct rpc_xprt *xprt,
1102                                        struct xdr_skb_reader *desc)
1103 {
1104         struct sock_xprt *transport =
1105                                 container_of(xprt, struct sock_xprt, xprt);
1106         struct rpc_rqst *req;
1107
1108         req = xprt_alloc_bc_request(xprt);
1109         if (req == NULL) {
1110                 printk(KERN_WARNING "Callback slot table overflowed\n");
1111                 xprt_force_disconnect(xprt);
1112                 return -1;
1113         }
1114
1115         req->rq_xid = transport->tcp_xid;
1116         dprintk("RPC:       read callback  XID %08x\n", ntohl(req->rq_xid));
1117         xs_tcp_read_common(xprt, desc, req);
1118
1119         if (!(transport->tcp_flags & TCP_RCV_COPY_DATA)) {
1120                 struct svc_serv *bc_serv = xprt->bc_serv;
1121
1122                 /*
1123                  * Add callback request to callback list.  The callback
1124                  * service sleeps on the sv_cb_waitq waiting for new
1125                  * requests.  Wake it up after adding enqueing the
1126                  * request.
1127                  */
1128                 dprintk("RPC:       add callback request to list\n");
1129                 spin_lock(&bc_serv->sv_cb_lock);
1130                 list_add(&req->rq_bc_list, &bc_serv->sv_cb_list);
1131                 spin_unlock(&bc_serv->sv_cb_lock);
1132                 wake_up(&bc_serv->sv_cb_waitq);
1133         }
1134
1135         req->rq_private_buf.len = transport->tcp_copied;
1136
1137         return 0;
1138 }
1139
1140 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1141                                         struct xdr_skb_reader *desc)
1142 {
1143         struct sock_xprt *transport =
1144                                 container_of(xprt, struct sock_xprt, xprt);
1145
1146         return (transport->tcp_flags & TCP_RPC_REPLY) ?
1147                 xs_tcp_read_reply(xprt, desc) :
1148                 xs_tcp_read_callback(xprt, desc);
1149 }
1150 #else
1151 static inline int _xs_tcp_read_data(struct rpc_xprt *xprt,
1152                                         struct xdr_skb_reader *desc)
1153 {
1154         return xs_tcp_read_reply(xprt, desc);
1155 }
1156 #endif /* CONFIG_NFS_V4_1 */
1157
1158 /*
1159  * Read data off the transport.  This can be either an RPC_CALL or an
1160  * RPC_REPLY.  Relay the processing to helper functions.
1161  */
1162 static void xs_tcp_read_data(struct rpc_xprt *xprt,
1163                                     struct xdr_skb_reader *desc)
1164 {
1165         struct sock_xprt *transport =
1166                                 container_of(xprt, struct sock_xprt, xprt);
1167
1168         if (_xs_tcp_read_data(xprt, desc) == 0)
1169                 xs_tcp_check_fraghdr(transport);
1170         else {
1171                 /*
1172                  * The transport_lock protects the request handling.
1173                  * There's no need to hold it to update the tcp_flags.
1174                  */
1175                 transport->tcp_flags &= ~TCP_RCV_COPY_DATA;
1176         }
1177 }
1178
1179 static inline void xs_tcp_read_discard(struct sock_xprt *transport, struct xdr_skb_reader *desc)
1180 {
1181         size_t len;
1182
1183         len = transport->tcp_reclen - transport->tcp_offset;
1184         if (len > desc->count)
1185                 len = desc->count;
1186         desc->count -= len;
1187         desc->offset += len;
1188         transport->tcp_offset += len;
1189         dprintk("RPC:       discarded %Zu bytes\n", len);
1190         xs_tcp_check_fraghdr(transport);
1191 }
1192
1193 static int xs_tcp_data_recv(read_descriptor_t *rd_desc, struct sk_buff *skb, unsigned int offset, size_t len)
1194 {
1195         struct rpc_xprt *xprt = rd_desc->arg.data;
1196         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1197         struct xdr_skb_reader desc = {
1198                 .skb    = skb,
1199                 .offset = offset,
1200                 .count  = len,
1201         };
1202
1203         dprintk("RPC:       xs_tcp_data_recv started\n");
1204         do {
1205                 /* Read in a new fragment marker if necessary */
1206                 /* Can we ever really expect to get completely empty fragments? */
1207                 if (transport->tcp_flags & TCP_RCV_COPY_FRAGHDR) {
1208                         xs_tcp_read_fraghdr(xprt, &desc);
1209                         continue;
1210                 }
1211                 /* Read in the xid if necessary */
1212                 if (transport->tcp_flags & TCP_RCV_COPY_XID) {
1213                         xs_tcp_read_xid(transport, &desc);
1214                         continue;
1215                 }
1216                 /* Read in the call/reply flag */
1217                 if (transport->tcp_flags & TCP_RCV_READ_CALLDIR) {
1218                         xs_tcp_read_calldir(transport, &desc);
1219                         continue;
1220                 }
1221                 /* Read in the request data */
1222                 if (transport->tcp_flags & TCP_RCV_COPY_DATA) {
1223                         xs_tcp_read_data(xprt, &desc);
1224                         continue;
1225                 }
1226                 /* Skip over any trailing bytes on short reads */
1227                 xs_tcp_read_discard(transport, &desc);
1228         } while (desc.count);
1229         dprintk("RPC:       xs_tcp_data_recv done\n");
1230         return len - desc.count;
1231 }
1232
1233 /**
1234  * xs_tcp_data_ready - "data ready" callback for TCP sockets
1235  * @sk: socket with data to read
1236  * @bytes: how much data to read
1237  *
1238  */
1239 static void xs_tcp_data_ready(struct sock *sk, int bytes)
1240 {
1241         struct rpc_xprt *xprt;
1242         read_descriptor_t rd_desc;
1243         int read;
1244
1245         dprintk("RPC:       xs_tcp_data_ready...\n");
1246
1247         read_lock(&sk->sk_callback_lock);
1248         if (!(xprt = xprt_from_sock(sk)))
1249                 goto out;
1250         if (xprt->shutdown)
1251                 goto out;
1252
1253         /* Any data means we had a useful conversation, so
1254          * the we don't need to delay the next reconnect
1255          */
1256         if (xprt->reestablish_timeout)
1257                 xprt->reestablish_timeout = 0;
1258
1259         /* We use rd_desc to pass struct xprt to xs_tcp_data_recv */
1260         rd_desc.arg.data = xprt;
1261         do {
1262                 rd_desc.count = 65536;
1263                 read = tcp_read_sock(sk, &rd_desc, xs_tcp_data_recv);
1264         } while (read > 0);
1265 out:
1266         read_unlock(&sk->sk_callback_lock);
1267 }
1268
1269 /*
1270  * Do the equivalent of linger/linger2 handling for dealing with
1271  * broken servers that don't close the socket in a timely
1272  * fashion
1273  */
1274 static void xs_tcp_schedule_linger_timeout(struct rpc_xprt *xprt,
1275                 unsigned long timeout)
1276 {
1277         struct sock_xprt *transport;
1278
1279         if (xprt_test_and_set_connecting(xprt))
1280                 return;
1281         set_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1282         transport = container_of(xprt, struct sock_xprt, xprt);
1283         queue_delayed_work(rpciod_workqueue, &transport->connect_worker,
1284                            timeout);
1285 }
1286
1287 static void xs_tcp_cancel_linger_timeout(struct rpc_xprt *xprt)
1288 {
1289         struct sock_xprt *transport;
1290
1291         transport = container_of(xprt, struct sock_xprt, xprt);
1292
1293         if (!test_bit(XPRT_CONNECTION_ABORT, &xprt->state) ||
1294             !cancel_delayed_work(&transport->connect_worker))
1295                 return;
1296         clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1297         xprt_clear_connecting(xprt);
1298 }
1299
1300 static void xs_sock_mark_closed(struct rpc_xprt *xprt)
1301 {
1302         smp_mb__before_clear_bit();
1303         clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1304         clear_bit(XPRT_CLOSING, &xprt->state);
1305         smp_mb__after_clear_bit();
1306         /* Mark transport as closed and wake up all pending tasks */
1307         xprt_disconnect_done(xprt);
1308 }
1309
1310 /**
1311  * xs_tcp_state_change - callback to handle TCP socket state changes
1312  * @sk: socket whose state has changed
1313  *
1314  */
1315 static void xs_tcp_state_change(struct sock *sk)
1316 {
1317         struct rpc_xprt *xprt;
1318
1319         read_lock(&sk->sk_callback_lock);
1320         if (!(xprt = xprt_from_sock(sk)))
1321                 goto out;
1322         dprintk("RPC:       xs_tcp_state_change client %p...\n", xprt);
1323         dprintk("RPC:       state %x conn %d dead %d zapped %d\n",
1324                         sk->sk_state, xprt_connected(xprt),
1325                         sock_flag(sk, SOCK_DEAD),
1326                         sock_flag(sk, SOCK_ZAPPED));
1327
1328         switch (sk->sk_state) {
1329         case TCP_ESTABLISHED:
1330                 spin_lock_bh(&xprt->transport_lock);
1331                 if (!xprt_test_and_set_connected(xprt)) {
1332                         struct sock_xprt *transport = container_of(xprt,
1333                                         struct sock_xprt, xprt);
1334
1335                         /* Reset TCP record info */
1336                         transport->tcp_offset = 0;
1337                         transport->tcp_reclen = 0;
1338                         transport->tcp_copied = 0;
1339                         transport->tcp_flags =
1340                                 TCP_RCV_COPY_FRAGHDR | TCP_RCV_COPY_XID;
1341
1342                         xprt_wake_pending_tasks(xprt, -EAGAIN);
1343                 }
1344                 spin_unlock_bh(&xprt->transport_lock);
1345                 break;
1346         case TCP_FIN_WAIT1:
1347                 /* The client initiated a shutdown of the socket */
1348                 xprt->connect_cookie++;
1349                 xprt->reestablish_timeout = 0;
1350                 set_bit(XPRT_CLOSING, &xprt->state);
1351                 smp_mb__before_clear_bit();
1352                 clear_bit(XPRT_CONNECTED, &xprt->state);
1353                 clear_bit(XPRT_CLOSE_WAIT, &xprt->state);
1354                 smp_mb__after_clear_bit();
1355                 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1356                 break;
1357         case TCP_CLOSE_WAIT:
1358                 /* The server initiated a shutdown of the socket */
1359                 xprt_force_disconnect(xprt);
1360         case TCP_SYN_SENT:
1361                 xprt->connect_cookie++;
1362         case TCP_CLOSING:
1363                 /*
1364                  * If the server closed down the connection, make sure that
1365                  * we back off before reconnecting
1366                  */
1367                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
1368                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
1369                 break;
1370         case TCP_LAST_ACK:
1371                 set_bit(XPRT_CLOSING, &xprt->state);
1372                 xs_tcp_schedule_linger_timeout(xprt, xs_tcp_fin_timeout);
1373                 smp_mb__before_clear_bit();
1374                 clear_bit(XPRT_CONNECTED, &xprt->state);
1375                 smp_mb__after_clear_bit();
1376                 break;
1377         case TCP_CLOSE:
1378                 xs_tcp_cancel_linger_timeout(xprt);
1379                 xs_sock_mark_closed(xprt);
1380         }
1381  out:
1382         read_unlock(&sk->sk_callback_lock);
1383 }
1384
1385 /**
1386  * xs_error_report - callback mainly for catching socket errors
1387  * @sk: socket
1388  */
1389 static void xs_error_report(struct sock *sk)
1390 {
1391         struct rpc_xprt *xprt;
1392
1393         read_lock(&sk->sk_callback_lock);
1394         if (!(xprt = xprt_from_sock(sk)))
1395                 goto out;
1396         dprintk("RPC:       %s client %p...\n"
1397                         "RPC:       error %d\n",
1398                         __func__, xprt, sk->sk_err);
1399         xprt_wake_pending_tasks(xprt, -EAGAIN);
1400 out:
1401         read_unlock(&sk->sk_callback_lock);
1402 }
1403
1404 static void xs_write_space(struct sock *sk)
1405 {
1406         struct socket *sock;
1407         struct rpc_xprt *xprt;
1408
1409         if (unlikely(!(sock = sk->sk_socket)))
1410                 return;
1411         clear_bit(SOCK_NOSPACE, &sock->flags);
1412
1413         if (unlikely(!(xprt = xprt_from_sock(sk))))
1414                 return;
1415         if (test_and_clear_bit(SOCK_ASYNC_NOSPACE, &sock->flags) == 0)
1416                 return;
1417
1418         xprt_write_space(xprt);
1419 }
1420
1421 /**
1422  * xs_udp_write_space - callback invoked when socket buffer space
1423  *                             becomes available
1424  * @sk: socket whose state has changed
1425  *
1426  * Called when more output buffer space is available for this socket.
1427  * We try not to wake our writers until they can make "significant"
1428  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1429  * with a bunch of small requests.
1430  */
1431 static void xs_udp_write_space(struct sock *sk)
1432 {
1433         read_lock(&sk->sk_callback_lock);
1434
1435         /* from net/core/sock.c:sock_def_write_space */
1436         if (sock_writeable(sk))
1437                 xs_write_space(sk);
1438
1439         read_unlock(&sk->sk_callback_lock);
1440 }
1441
1442 /**
1443  * xs_tcp_write_space - callback invoked when socket buffer space
1444  *                             becomes available
1445  * @sk: socket whose state has changed
1446  *
1447  * Called when more output buffer space is available for this socket.
1448  * We try not to wake our writers until they can make "significant"
1449  * progress, otherwise we'll waste resources thrashing kernel_sendmsg
1450  * with a bunch of small requests.
1451  */
1452 static void xs_tcp_write_space(struct sock *sk)
1453 {
1454         read_lock(&sk->sk_callback_lock);
1455
1456         /* from net/core/stream.c:sk_stream_write_space */
1457         if (sk_stream_wspace(sk) >= sk_stream_min_wspace(sk))
1458                 xs_write_space(sk);
1459
1460         read_unlock(&sk->sk_callback_lock);
1461 }
1462
1463 static void xs_udp_do_set_buffer_size(struct rpc_xprt *xprt)
1464 {
1465         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1466         struct sock *sk = transport->inet;
1467
1468         if (transport->rcvsize) {
1469                 sk->sk_userlocks |= SOCK_RCVBUF_LOCK;
1470                 sk->sk_rcvbuf = transport->rcvsize * xprt->max_reqs * 2;
1471         }
1472         if (transport->sndsize) {
1473                 sk->sk_userlocks |= SOCK_SNDBUF_LOCK;
1474                 sk->sk_sndbuf = transport->sndsize * xprt->max_reqs * 2;
1475                 sk->sk_write_space(sk);
1476         }
1477 }
1478
1479 /**
1480  * xs_udp_set_buffer_size - set send and receive limits
1481  * @xprt: generic transport
1482  * @sndsize: requested size of send buffer, in bytes
1483  * @rcvsize: requested size of receive buffer, in bytes
1484  *
1485  * Set socket send and receive buffer size limits.
1486  */
1487 static void xs_udp_set_buffer_size(struct rpc_xprt *xprt, size_t sndsize, size_t rcvsize)
1488 {
1489         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1490
1491         transport->sndsize = 0;
1492         if (sndsize)
1493                 transport->sndsize = sndsize + 1024;
1494         transport->rcvsize = 0;
1495         if (rcvsize)
1496                 transport->rcvsize = rcvsize + 1024;
1497
1498         xs_udp_do_set_buffer_size(xprt);
1499 }
1500
1501 /**
1502  * xs_udp_timer - called when a retransmit timeout occurs on a UDP transport
1503  * @task: task that timed out
1504  *
1505  * Adjust the congestion window after a retransmit timeout has occurred.
1506  */
1507 static void xs_udp_timer(struct rpc_task *task)
1508 {
1509         xprt_adjust_cwnd(task, -ETIMEDOUT);
1510 }
1511
1512 static unsigned short xs_get_random_port(void)
1513 {
1514         unsigned short range = xprt_max_resvport - xprt_min_resvport;
1515         unsigned short rand = (unsigned short) net_random() % range;
1516         return rand + xprt_min_resvport;
1517 }
1518
1519 /**
1520  * xs_set_port - reset the port number in the remote endpoint address
1521  * @xprt: generic transport
1522  * @port: new port number
1523  *
1524  */
1525 static void xs_set_port(struct rpc_xprt *xprt, unsigned short port)
1526 {
1527         dprintk("RPC:       setting port for xprt %p to %u\n", xprt, port);
1528
1529         rpc_set_port(xs_addr(xprt), port);
1530         xs_update_peer_port(xprt);
1531 }
1532
1533 static unsigned short xs_get_srcport(struct sock_xprt *transport, struct socket *sock)
1534 {
1535         unsigned short port = transport->srcport;
1536
1537         if (port == 0 && transport->xprt.resvport)
1538                 port = xs_get_random_port();
1539         return port;
1540 }
1541
1542 static unsigned short xs_next_srcport(struct sock_xprt *transport, struct socket *sock, unsigned short port)
1543 {
1544         if (transport->srcport != 0)
1545                 transport->srcport = 0;
1546         if (!transport->xprt.resvport)
1547                 return 0;
1548         if (port <= xprt_min_resvport || port > xprt_max_resvport)
1549                 return xprt_max_resvport;
1550         return --port;
1551 }
1552
1553 static int xs_bind4(struct sock_xprt *transport, struct socket *sock)
1554 {
1555         struct sockaddr_in myaddr = {
1556                 .sin_family = AF_INET,
1557         };
1558         struct sockaddr_in *sa;
1559         int err, nloop = 0;
1560         unsigned short port = xs_get_srcport(transport, sock);
1561         unsigned short last;
1562
1563         sa = (struct sockaddr_in *)&transport->srcaddr;
1564         myaddr.sin_addr = sa->sin_addr;
1565         do {
1566                 myaddr.sin_port = htons(port);
1567                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1568                                                 sizeof(myaddr));
1569                 if (port == 0)
1570                         break;
1571                 if (err == 0) {
1572                         transport->srcport = port;
1573                         break;
1574                 }
1575                 last = port;
1576                 port = xs_next_srcport(transport, sock, port);
1577                 if (port > last)
1578                         nloop++;
1579         } while (err == -EADDRINUSE && nloop != 2);
1580         dprintk("RPC:       %s %pI4:%u: %s (%d)\n",
1581                         __func__, &myaddr.sin_addr,
1582                         port, err ? "failed" : "ok", err);
1583         return err;
1584 }
1585
1586 static int xs_bind6(struct sock_xprt *transport, struct socket *sock)
1587 {
1588         struct sockaddr_in6 myaddr = {
1589                 .sin6_family = AF_INET6,
1590         };
1591         struct sockaddr_in6 *sa;
1592         int err, nloop = 0;
1593         unsigned short port = xs_get_srcport(transport, sock);
1594         unsigned short last;
1595
1596         sa = (struct sockaddr_in6 *)&transport->srcaddr;
1597         myaddr.sin6_addr = sa->sin6_addr;
1598         do {
1599                 myaddr.sin6_port = htons(port);
1600                 err = kernel_bind(sock, (struct sockaddr *) &myaddr,
1601                                                 sizeof(myaddr));
1602                 if (port == 0)
1603                         break;
1604                 if (err == 0) {
1605                         transport->srcport = port;
1606                         break;
1607                 }
1608                 last = port;
1609                 port = xs_next_srcport(transport, sock, port);
1610                 if (port > last)
1611                         nloop++;
1612         } while (err == -EADDRINUSE && nloop != 2);
1613         dprintk("RPC:       xs_bind6 %pI6:%u: %s (%d)\n",
1614                 &myaddr.sin6_addr, port, err ? "failed" : "ok", err);
1615         return err;
1616 }
1617
1618 #ifdef CONFIG_DEBUG_LOCK_ALLOC
1619 static struct lock_class_key xs_key[2];
1620 static struct lock_class_key xs_slock_key[2];
1621
1622 static inline void xs_reclassify_socket4(struct socket *sock)
1623 {
1624         struct sock *sk = sock->sk;
1625
1626         BUG_ON(sock_owned_by_user(sk));
1627         sock_lock_init_class_and_name(sk, "slock-AF_INET-RPC",
1628                 &xs_slock_key[0], "sk_lock-AF_INET-RPC", &xs_key[0]);
1629 }
1630
1631 static inline void xs_reclassify_socket6(struct socket *sock)
1632 {
1633         struct sock *sk = sock->sk;
1634
1635         BUG_ON(sock_owned_by_user(sk));
1636         sock_lock_init_class_and_name(sk, "slock-AF_INET6-RPC",
1637                 &xs_slock_key[1], "sk_lock-AF_INET6-RPC", &xs_key[1]);
1638 }
1639 #else
1640 static inline void xs_reclassify_socket4(struct socket *sock)
1641 {
1642 }
1643
1644 static inline void xs_reclassify_socket6(struct socket *sock)
1645 {
1646 }
1647 #endif
1648
1649 static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1650 {
1651         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1652
1653         if (!transport->inet) {
1654                 struct sock *sk = sock->sk;
1655
1656                 write_lock_bh(&sk->sk_callback_lock);
1657
1658                 xs_save_old_callbacks(transport, sk);
1659
1660                 sk->sk_user_data = xprt;
1661                 sk->sk_data_ready = xs_udp_data_ready;
1662                 sk->sk_write_space = xs_udp_write_space;
1663                 sk->sk_error_report = xs_error_report;
1664                 sk->sk_no_check = UDP_CSUM_NORCV;
1665                 sk->sk_allocation = GFP_ATOMIC;
1666
1667                 xprt_set_connected(xprt);
1668
1669                 /* Reset to new socket */
1670                 transport->sock = sock;
1671                 transport->inet = sk;
1672
1673                 write_unlock_bh(&sk->sk_callback_lock);
1674         }
1675         xs_udp_do_set_buffer_size(xprt);
1676 }
1677
1678 /**
1679  * xs_udp_connect_worker4 - set up a UDP socket
1680  * @work: RPC transport to connect
1681  *
1682  * Invoked by a work queue tasklet.
1683  */
1684 static void xs_udp_connect_worker4(struct work_struct *work)
1685 {
1686         struct sock_xprt *transport =
1687                 container_of(work, struct sock_xprt, connect_worker.work);
1688         struct rpc_xprt *xprt = &transport->xprt;
1689         struct socket *sock = transport->sock;
1690         int err, status = -EIO;
1691
1692         if (xprt->shutdown)
1693                 goto out;
1694
1695         /* Start by resetting any existing state */
1696         xs_reset_transport(transport);
1697
1698         err = sock_create_kern(PF_INET, SOCK_DGRAM, IPPROTO_UDP, &sock);
1699         if (err < 0) {
1700                 dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1701                 goto out;
1702         }
1703         xs_reclassify_socket4(sock);
1704
1705         if (xs_bind4(transport, sock)) {
1706                 sock_release(sock);
1707                 goto out;
1708         }
1709
1710         dprintk("RPC:       worker connecting xprt %p via %s to "
1711                                 "%s (port %s)\n", xprt,
1712                         xprt->address_strings[RPC_DISPLAY_PROTO],
1713                         xprt->address_strings[RPC_DISPLAY_ADDR],
1714                         xprt->address_strings[RPC_DISPLAY_PORT]);
1715
1716         xs_udp_finish_connecting(xprt, sock);
1717         status = 0;
1718 out:
1719         xprt_clear_connecting(xprt);
1720         xprt_wake_pending_tasks(xprt, status);
1721 }
1722
1723 /**
1724  * xs_udp_connect_worker6 - set up a UDP socket
1725  * @work: RPC transport to connect
1726  *
1727  * Invoked by a work queue tasklet.
1728  */
1729 static void xs_udp_connect_worker6(struct work_struct *work)
1730 {
1731         struct sock_xprt *transport =
1732                 container_of(work, struct sock_xprt, connect_worker.work);
1733         struct rpc_xprt *xprt = &transport->xprt;
1734         struct socket *sock = transport->sock;
1735         int err, status = -EIO;
1736
1737         if (xprt->shutdown)
1738                 goto out;
1739
1740         /* Start by resetting any existing state */
1741         xs_reset_transport(transport);
1742
1743         err = sock_create_kern(PF_INET6, SOCK_DGRAM, IPPROTO_UDP, &sock);
1744         if (err < 0) {
1745                 dprintk("RPC:       can't create UDP transport socket (%d).\n", -err);
1746                 goto out;
1747         }
1748         xs_reclassify_socket6(sock);
1749
1750         if (xs_bind6(transport, sock) < 0) {
1751                 sock_release(sock);
1752                 goto out;
1753         }
1754
1755         dprintk("RPC:       worker connecting xprt %p via %s to "
1756                                 "%s (port %s)\n", xprt,
1757                         xprt->address_strings[RPC_DISPLAY_PROTO],
1758                         xprt->address_strings[RPC_DISPLAY_ADDR],
1759                         xprt->address_strings[RPC_DISPLAY_PORT]);
1760
1761         xs_udp_finish_connecting(xprt, sock);
1762         status = 0;
1763 out:
1764         xprt_clear_connecting(xprt);
1765         xprt_wake_pending_tasks(xprt, status);
1766 }
1767
1768 /*
1769  * We need to preserve the port number so the reply cache on the server can
1770  * find our cached RPC replies when we get around to reconnecting.
1771  */
1772 static void xs_abort_connection(struct rpc_xprt *xprt, struct sock_xprt *transport)
1773 {
1774         int result;
1775         struct sockaddr any;
1776
1777         dprintk("RPC:       disconnecting xprt %p to reuse port\n", xprt);
1778
1779         /*
1780          * Disconnect the transport socket by doing a connect operation
1781          * with AF_UNSPEC.  This should return immediately...
1782          */
1783         memset(&any, 0, sizeof(any));
1784         any.sa_family = AF_UNSPEC;
1785         result = kernel_connect(transport->sock, &any, sizeof(any), 0);
1786         if (!result)
1787                 xs_sock_mark_closed(xprt);
1788         else
1789                 dprintk("RPC:       AF_UNSPEC connect return code %d\n",
1790                                 result);
1791 }
1792
1793 static void xs_tcp_reuse_connection(struct rpc_xprt *xprt, struct sock_xprt *transport)
1794 {
1795         unsigned int state = transport->inet->sk_state;
1796
1797         if (state == TCP_CLOSE && transport->sock->state == SS_UNCONNECTED)
1798                 return;
1799         if ((1 << state) & (TCPF_ESTABLISHED|TCPF_SYN_SENT))
1800                 return;
1801         xs_abort_connection(xprt, transport);
1802 }
1803
1804 static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
1805 {
1806         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
1807
1808         if (!transport->inet) {
1809                 struct sock *sk = sock->sk;
1810
1811                 write_lock_bh(&sk->sk_callback_lock);
1812
1813                 xs_save_old_callbacks(transport, sk);
1814
1815                 sk->sk_user_data = xprt;
1816                 sk->sk_data_ready = xs_tcp_data_ready;
1817                 sk->sk_state_change = xs_tcp_state_change;
1818                 sk->sk_write_space = xs_tcp_write_space;
1819                 sk->sk_error_report = xs_error_report;
1820                 sk->sk_allocation = GFP_ATOMIC;
1821
1822                 /* socket options */
1823                 sk->sk_userlocks |= SOCK_BINDPORT_LOCK;
1824                 sock_reset_flag(sk, SOCK_LINGER);
1825                 tcp_sk(sk)->linger2 = 0;
1826                 tcp_sk(sk)->nonagle |= TCP_NAGLE_OFF;
1827
1828                 xprt_clear_connected(xprt);
1829
1830                 /* Reset to new socket */
1831                 transport->sock = sock;
1832                 transport->inet = sk;
1833
1834                 write_unlock_bh(&sk->sk_callback_lock);
1835         }
1836
1837         if (!xprt_bound(xprt))
1838                 return -ENOTCONN;
1839
1840         /* Tell the socket layer to start connecting... */
1841         xprt->stat.connect_count++;
1842         xprt->stat.connect_start = jiffies;
1843         return kernel_connect(sock, xs_addr(xprt), xprt->addrlen, O_NONBLOCK);
1844 }
1845
1846 /**
1847  * xs_tcp_setup_socket - create a TCP socket and connect to a remote endpoint
1848  * @xprt: RPC transport to connect
1849  * @transport: socket transport to connect
1850  * @create_sock: function to create a socket of the correct type
1851  *
1852  * Invoked by a work queue tasklet.
1853  */
1854 static void xs_tcp_setup_socket(struct rpc_xprt *xprt,
1855                 struct sock_xprt *transport,
1856                 struct socket *(*create_sock)(struct rpc_xprt *,
1857                         struct sock_xprt *))
1858 {
1859         struct socket *sock = transport->sock;
1860         int status = -EIO;
1861
1862         if (xprt->shutdown)
1863                 goto out;
1864
1865         if (!sock) {
1866                 clear_bit(XPRT_CONNECTION_ABORT, &xprt->state);
1867                 sock = create_sock(xprt, transport);
1868                 if (IS_ERR(sock)) {
1869                         status = PTR_ERR(sock);
1870                         goto out;
1871                 }
1872         } else {
1873                 int abort_and_exit;
1874
1875                 abort_and_exit = test_and_clear_bit(XPRT_CONNECTION_ABORT,
1876                                 &xprt->state);
1877                 /* "close" the socket, preserving the local port */
1878                 xs_tcp_reuse_connection(xprt, transport);
1879
1880                 if (abort_and_exit)
1881                         goto out_eagain;
1882         }
1883
1884         dprintk("RPC:       worker connecting xprt %p via %s to "
1885                                 "%s (port %s)\n", xprt,
1886                         xprt->address_strings[RPC_DISPLAY_PROTO],
1887                         xprt->address_strings[RPC_DISPLAY_ADDR],
1888                         xprt->address_strings[RPC_DISPLAY_PORT]);
1889
1890         status = xs_tcp_finish_connecting(xprt, sock);
1891         dprintk("RPC:       %p connect status %d connected %d sock state %d\n",
1892                         xprt, -status, xprt_connected(xprt),
1893                         sock->sk->sk_state);
1894         switch (status) {
1895         default:
1896                 printk("%s: connect returned unhandled error %d\n",
1897                         __func__, status);
1898         case -EADDRNOTAVAIL:
1899                 /* We're probably in TIME_WAIT. Get rid of existing socket,
1900                  * and retry
1901                  */
1902                 set_bit(XPRT_CONNECTION_CLOSE, &xprt->state);
1903                 xprt_force_disconnect(xprt);
1904                 break;
1905         case -ECONNREFUSED:
1906         case -ECONNRESET:
1907         case -ENETUNREACH:
1908                 /* retry with existing socket, after a delay */
1909         case 0:
1910         case -EINPROGRESS:
1911         case -EALREADY:
1912                 xprt_clear_connecting(xprt);
1913                 return;
1914         }
1915 out_eagain:
1916         status = -EAGAIN;
1917 out:
1918         xprt_clear_connecting(xprt);
1919         xprt_wake_pending_tasks(xprt, status);
1920 }
1921
1922 static struct socket *xs_create_tcp_sock4(struct rpc_xprt *xprt,
1923                 struct sock_xprt *transport)
1924 {
1925         struct socket *sock;
1926         int err;
1927
1928         /* start from scratch */
1929         err = sock_create_kern(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock);
1930         if (err < 0) {
1931                 dprintk("RPC:       can't create TCP transport socket (%d).\n",
1932                                 -err);
1933                 goto out_err;
1934         }
1935         xs_reclassify_socket4(sock);
1936
1937         if (xs_bind4(transport, sock) < 0) {
1938                 sock_release(sock);
1939                 goto out_err;
1940         }
1941         return sock;
1942 out_err:
1943         return ERR_PTR(-EIO);
1944 }
1945
1946 /**
1947  * xs_tcp_connect_worker4 - connect a TCP socket to a remote endpoint
1948  * @work: RPC transport to connect
1949  *
1950  * Invoked by a work queue tasklet.
1951  */
1952 static void xs_tcp_connect_worker4(struct work_struct *work)
1953 {
1954         struct sock_xprt *transport =
1955                 container_of(work, struct sock_xprt, connect_worker.work);
1956         struct rpc_xprt *xprt = &transport->xprt;
1957
1958         xs_tcp_setup_socket(xprt, transport, xs_create_tcp_sock4);
1959 }
1960
1961 static struct socket *xs_create_tcp_sock6(struct rpc_xprt *xprt,
1962                 struct sock_xprt *transport)
1963 {
1964         struct socket *sock;
1965         int err;
1966
1967         /* start from scratch */
1968         err = sock_create_kern(PF_INET6, SOCK_STREAM, IPPROTO_TCP, &sock);
1969         if (err < 0) {
1970                 dprintk("RPC:       can't create TCP transport socket (%d).\n",
1971                                 -err);
1972                 goto out_err;
1973         }
1974         xs_reclassify_socket6(sock);
1975
1976         if (xs_bind6(transport, sock) < 0) {
1977                 sock_release(sock);
1978                 goto out_err;
1979         }
1980         return sock;
1981 out_err:
1982         return ERR_PTR(-EIO);
1983 }
1984
1985 /**
1986  * xs_tcp_connect_worker6 - connect a TCP socket to a remote endpoint
1987  * @work: RPC transport to connect
1988  *
1989  * Invoked by a work queue tasklet.
1990  */
1991 static void xs_tcp_connect_worker6(struct work_struct *work)
1992 {
1993         struct sock_xprt *transport =
1994                 container_of(work, struct sock_xprt, connect_worker.work);
1995         struct rpc_xprt *xprt = &transport->xprt;
1996
1997         xs_tcp_setup_socket(xprt, transport, xs_create_tcp_sock6);
1998 }
1999
2000 /**
2001  * xs_connect - connect a socket to a remote endpoint
2002  * @task: address of RPC task that manages state of connect request
2003  *
2004  * TCP: If the remote end dropped the connection, delay reconnecting.
2005  *
2006  * UDP socket connects are synchronous, but we use a work queue anyway
2007  * to guarantee that even unprivileged user processes can set up a
2008  * socket on a privileged port.
2009  *
2010  * If a UDP socket connect fails, the delay behavior here prevents
2011  * retry floods (hard mounts).
2012  */
2013 static void xs_connect(struct rpc_task *task)
2014 {
2015         struct rpc_xprt *xprt = task->tk_xprt;
2016         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2017
2018         if (xprt_test_and_set_connecting(xprt))
2019                 return;
2020
2021         if (transport->sock != NULL && !RPC_IS_SOFTCONN(task)) {
2022                 dprintk("RPC:       xs_connect delayed xprt %p for %lu "
2023                                 "seconds\n",
2024                                 xprt, xprt->reestablish_timeout / HZ);
2025                 queue_delayed_work(rpciod_workqueue,
2026                                    &transport->connect_worker,
2027                                    xprt->reestablish_timeout);
2028                 xprt->reestablish_timeout <<= 1;
2029                 if (xprt->reestablish_timeout < XS_TCP_INIT_REEST_TO)
2030                         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2031                 if (xprt->reestablish_timeout > XS_TCP_MAX_REEST_TO)
2032                         xprt->reestablish_timeout = XS_TCP_MAX_REEST_TO;
2033         } else {
2034                 dprintk("RPC:       xs_connect scheduled xprt %p\n", xprt);
2035                 queue_delayed_work(rpciod_workqueue,
2036                                    &transport->connect_worker, 0);
2037         }
2038 }
2039
2040 static void xs_tcp_connect(struct rpc_task *task)
2041 {
2042         struct rpc_xprt *xprt = task->tk_xprt;
2043
2044         /* Exit if we need to wait for socket shutdown to complete */
2045         if (test_bit(XPRT_CLOSING, &xprt->state))
2046                 return;
2047         xs_connect(task);
2048 }
2049
2050 /**
2051  * xs_udp_print_stats - display UDP socket-specifc stats
2052  * @xprt: rpc_xprt struct containing statistics
2053  * @seq: output file
2054  *
2055  */
2056 static void xs_udp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2057 {
2058         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2059
2060         seq_printf(seq, "\txprt:\tudp %u %lu %lu %lu %lu %Lu %Lu\n",
2061                         transport->srcport,
2062                         xprt->stat.bind_count,
2063                         xprt->stat.sends,
2064                         xprt->stat.recvs,
2065                         xprt->stat.bad_xids,
2066                         xprt->stat.req_u,
2067                         xprt->stat.bklog_u);
2068 }
2069
2070 /**
2071  * xs_tcp_print_stats - display TCP socket-specifc stats
2072  * @xprt: rpc_xprt struct containing statistics
2073  * @seq: output file
2074  *
2075  */
2076 static void xs_tcp_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
2077 {
2078         struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
2079         long idle_time = 0;
2080
2081         if (xprt_connected(xprt))
2082                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
2083
2084         seq_printf(seq, "\txprt:\ttcp %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu\n",
2085                         transport->srcport,
2086                         xprt->stat.bind_count,
2087                         xprt->stat.connect_count,
2088                         xprt->stat.connect_time,
2089                         idle_time,
2090                         xprt->stat.sends,
2091                         xprt->stat.recvs,
2092                         xprt->stat.bad_xids,
2093                         xprt->stat.req_u,
2094                         xprt->stat.bklog_u);
2095 }
2096
2097 /*
2098  * Allocate a bunch of pages for a scratch buffer for the rpc code. The reason
2099  * we allocate pages instead doing a kmalloc like rpc_malloc is because we want
2100  * to use the server side send routines.
2101  */
2102 void *bc_malloc(struct rpc_task *task, size_t size)
2103 {
2104         struct page *page;
2105         struct rpc_buffer *buf;
2106
2107         BUG_ON(size > PAGE_SIZE - sizeof(struct rpc_buffer));
2108         page = alloc_page(GFP_KERNEL);
2109
2110         if (!page)
2111                 return NULL;
2112
2113         buf = page_address(page);
2114         buf->len = PAGE_SIZE;
2115
2116         return buf->data;
2117 }
2118
2119 /*
2120  * Free the space allocated in the bc_alloc routine
2121  */
2122 void bc_free(void *buffer)
2123 {
2124         struct rpc_buffer *buf;
2125
2126         if (!buffer)
2127                 return;
2128
2129         buf = container_of(buffer, struct rpc_buffer, data);
2130         free_page((unsigned long)buf);
2131 }
2132
2133 /*
2134  * Use the svc_sock to send the callback. Must be called with svsk->sk_mutex
2135  * held. Borrows heavily from svc_tcp_sendto and xs_tcp_send_request.
2136  */
2137 static int bc_sendto(struct rpc_rqst *req)
2138 {
2139         int len;
2140         struct xdr_buf *xbufp = &req->rq_snd_buf;
2141         struct rpc_xprt *xprt = req->rq_xprt;
2142         struct sock_xprt *transport =
2143                                 container_of(xprt, struct sock_xprt, xprt);
2144         struct socket *sock = transport->sock;
2145         unsigned long headoff;
2146         unsigned long tailoff;
2147
2148         /*
2149          * Set up the rpc header and record marker stuff
2150          */
2151         xs_encode_tcp_record_marker(xbufp);
2152
2153         tailoff = (unsigned long)xbufp->tail[0].iov_base & ~PAGE_MASK;
2154         headoff = (unsigned long)xbufp->head[0].iov_base & ~PAGE_MASK;
2155         len = svc_send_common(sock, xbufp,
2156                               virt_to_page(xbufp->head[0].iov_base), headoff,
2157                               xbufp->tail[0].iov_base, tailoff);
2158
2159         if (len != xbufp->len) {
2160                 printk(KERN_NOTICE "Error sending entire callback!\n");
2161                 len = -EAGAIN;
2162         }
2163
2164         return len;
2165 }
2166
2167 /*
2168  * The send routine. Borrows from svc_send
2169  */
2170 static int bc_send_request(struct rpc_task *task)
2171 {
2172         struct rpc_rqst *req = task->tk_rqstp;
2173         struct svc_xprt *xprt;
2174         struct svc_sock         *svsk;
2175         u32                     len;
2176
2177         dprintk("sending request with xid: %08x\n", ntohl(req->rq_xid));
2178         /*
2179          * Get the server socket associated with this callback xprt
2180          */
2181         xprt = req->rq_xprt->bc_xprt;
2182         svsk = container_of(xprt, struct svc_sock, sk_xprt);
2183
2184         /*
2185          * Grab the mutex to serialize data as the connection is shared
2186          * with the fore channel
2187          */
2188         if (!mutex_trylock(&xprt->xpt_mutex)) {
2189                 rpc_sleep_on(&xprt->xpt_bc_pending, task, NULL);
2190                 if (!mutex_trylock(&xprt->xpt_mutex))
2191                         return -EAGAIN;
2192                 rpc_wake_up_queued_task(&xprt->xpt_bc_pending, task);
2193         }
2194         if (test_bit(XPT_DEAD, &xprt->xpt_flags))
2195                 len = -ENOTCONN;
2196         else
2197                 len = bc_sendto(req);
2198         mutex_unlock(&xprt->xpt_mutex);
2199
2200         if (len > 0)
2201                 len = 0;
2202
2203         return len;
2204 }
2205
2206 /*
2207  * The close routine. Since this is client initiated, we do nothing
2208  */
2209
2210 static void bc_close(struct rpc_xprt *xprt)
2211 {
2212         return;
2213 }
2214
2215 /*
2216  * The xprt destroy routine. Again, because this connection is client
2217  * initiated, we do nothing
2218  */
2219
2220 static void bc_destroy(struct rpc_xprt *xprt)
2221 {
2222         return;
2223 }
2224
2225 static struct rpc_xprt_ops xs_udp_ops = {
2226         .set_buffer_size        = xs_udp_set_buffer_size,
2227         .reserve_xprt           = xprt_reserve_xprt_cong,
2228         .release_xprt           = xprt_release_xprt_cong,
2229         .rpcbind                = rpcb_getport_async,
2230         .set_port               = xs_set_port,
2231         .connect                = xs_connect,
2232         .buf_alloc              = rpc_malloc,
2233         .buf_free               = rpc_free,
2234         .send_request           = xs_udp_send_request,
2235         .set_retrans_timeout    = xprt_set_retrans_timeout_rtt,
2236         .timer                  = xs_udp_timer,
2237         .release_request        = xprt_release_rqst_cong,
2238         .close                  = xs_close,
2239         .destroy                = xs_destroy,
2240         .print_stats            = xs_udp_print_stats,
2241 };
2242
2243 static struct rpc_xprt_ops xs_tcp_ops = {
2244         .reserve_xprt           = xprt_reserve_xprt,
2245         .release_xprt           = xs_tcp_release_xprt,
2246         .rpcbind                = rpcb_getport_async,
2247         .set_port               = xs_set_port,
2248         .connect                = xs_tcp_connect,
2249         .buf_alloc              = rpc_malloc,
2250         .buf_free               = rpc_free,
2251         .send_request           = xs_tcp_send_request,
2252         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2253 #if defined(CONFIG_NFS_V4_1)
2254         .release_request        = bc_release_request,
2255 #endif /* CONFIG_NFS_V4_1 */
2256         .close                  = xs_tcp_close,
2257         .destroy                = xs_destroy,
2258         .print_stats            = xs_tcp_print_stats,
2259 };
2260
2261 /*
2262  * The rpc_xprt_ops for the server backchannel
2263  */
2264
2265 static struct rpc_xprt_ops bc_tcp_ops = {
2266         .reserve_xprt           = xprt_reserve_xprt,
2267         .release_xprt           = xprt_release_xprt,
2268         .buf_alloc              = bc_malloc,
2269         .buf_free               = bc_free,
2270         .send_request           = bc_send_request,
2271         .set_retrans_timeout    = xprt_set_retrans_timeout_def,
2272         .close                  = bc_close,
2273         .destroy                = bc_destroy,
2274         .print_stats            = xs_tcp_print_stats,
2275 };
2276
2277 static struct rpc_xprt *xs_setup_xprt(struct xprt_create *args,
2278                                       unsigned int slot_table_size)
2279 {
2280         struct rpc_xprt *xprt;
2281         struct sock_xprt *new;
2282
2283         if (args->addrlen > sizeof(xprt->addr)) {
2284                 dprintk("RPC:       xs_setup_xprt: address too large\n");
2285                 return ERR_PTR(-EBADF);
2286         }
2287
2288         new = kzalloc(sizeof(*new), GFP_KERNEL);
2289         if (new == NULL) {
2290                 dprintk("RPC:       xs_setup_xprt: couldn't allocate "
2291                                 "rpc_xprt\n");
2292                 return ERR_PTR(-ENOMEM);
2293         }
2294         xprt = &new->xprt;
2295
2296         xprt->max_reqs = slot_table_size;
2297         xprt->slot = kcalloc(xprt->max_reqs, sizeof(struct rpc_rqst), GFP_KERNEL);
2298         if (xprt->slot == NULL) {
2299                 kfree(xprt);
2300                 dprintk("RPC:       xs_setup_xprt: couldn't allocate slot "
2301                                 "table\n");
2302                 return ERR_PTR(-ENOMEM);
2303         }
2304
2305         memcpy(&xprt->addr, args->dstaddr, args->addrlen);
2306         xprt->addrlen = args->addrlen;
2307         if (args->srcaddr)
2308                 memcpy(&new->srcaddr, args->srcaddr, args->addrlen);
2309
2310         return xprt;
2311 }
2312
2313 static const struct rpc_timeout xs_udp_default_timeout = {
2314         .to_initval = 5 * HZ,
2315         .to_maxval = 30 * HZ,
2316         .to_increment = 5 * HZ,
2317         .to_retries = 5,
2318 };
2319
2320 /**
2321  * xs_setup_udp - Set up transport to use a UDP socket
2322  * @args: rpc transport creation arguments
2323  *
2324  */
2325 static struct rpc_xprt *xs_setup_udp(struct xprt_create *args)
2326 {
2327         struct sockaddr *addr = args->dstaddr;
2328         struct rpc_xprt *xprt;
2329         struct sock_xprt *transport;
2330
2331         xprt = xs_setup_xprt(args, xprt_udp_slot_table_entries);
2332         if (IS_ERR(xprt))
2333                 return xprt;
2334         transport = container_of(xprt, struct sock_xprt, xprt);
2335
2336         xprt->prot = IPPROTO_UDP;
2337         xprt->tsh_size = 0;
2338         /* XXX: header size can vary due to auth type, IPv6, etc. */
2339         xprt->max_payload = (1U << 16) - (MAX_HEADER << 3);
2340
2341         xprt->bind_timeout = XS_BIND_TO;
2342         xprt->connect_timeout = XS_UDP_CONN_TO;
2343         xprt->reestablish_timeout = XS_UDP_REEST_TO;
2344         xprt->idle_timeout = XS_IDLE_DISC_TO;
2345
2346         xprt->ops = &xs_udp_ops;
2347
2348         xprt->timeout = &xs_udp_default_timeout;
2349
2350         switch (addr->sa_family) {
2351         case AF_INET:
2352                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2353                         xprt_set_bound(xprt);
2354
2355                 INIT_DELAYED_WORK(&transport->connect_worker,
2356                                         xs_udp_connect_worker4);
2357                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP);
2358                 break;
2359         case AF_INET6:
2360                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2361                         xprt_set_bound(xprt);
2362
2363                 INIT_DELAYED_WORK(&transport->connect_worker,
2364                                         xs_udp_connect_worker6);
2365                 xs_format_peer_addresses(xprt, "udp", RPCBIND_NETID_UDP6);
2366                 break;
2367         default:
2368                 kfree(xprt);
2369                 return ERR_PTR(-EAFNOSUPPORT);
2370         }
2371
2372         if (xprt_bound(xprt))
2373                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2374                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2375                                 xprt->address_strings[RPC_DISPLAY_PORT],
2376                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2377         else
2378                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2379                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2380                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2381
2382         if (try_module_get(THIS_MODULE))
2383                 return xprt;
2384
2385         kfree(xprt->slot);
2386         kfree(xprt);
2387         return ERR_PTR(-EINVAL);
2388 }
2389
2390 static const struct rpc_timeout xs_tcp_default_timeout = {
2391         .to_initval = 60 * HZ,
2392         .to_maxval = 60 * HZ,
2393         .to_retries = 2,
2394 };
2395
2396 /**
2397  * xs_setup_tcp - Set up transport to use a TCP socket
2398  * @args: rpc transport creation arguments
2399  *
2400  */
2401 static struct rpc_xprt *xs_setup_tcp(struct xprt_create *args)
2402 {
2403         struct sockaddr *addr = args->dstaddr;
2404         struct rpc_xprt *xprt;
2405         struct sock_xprt *transport;
2406
2407         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
2408         if (IS_ERR(xprt))
2409                 return xprt;
2410         transport = container_of(xprt, struct sock_xprt, xprt);
2411
2412         xprt->prot = IPPROTO_TCP;
2413         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2414         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2415
2416         xprt->bind_timeout = XS_BIND_TO;
2417         xprt->connect_timeout = XS_TCP_CONN_TO;
2418         xprt->reestablish_timeout = XS_TCP_INIT_REEST_TO;
2419         xprt->idle_timeout = XS_IDLE_DISC_TO;
2420
2421         xprt->ops = &xs_tcp_ops;
2422         xprt->timeout = &xs_tcp_default_timeout;
2423
2424         switch (addr->sa_family) {
2425         case AF_INET:
2426                 if (((struct sockaddr_in *)addr)->sin_port != htons(0))
2427                         xprt_set_bound(xprt);
2428
2429                 INIT_DELAYED_WORK(&transport->connect_worker,
2430                                         xs_tcp_connect_worker4);
2431                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP);
2432                 break;
2433         case AF_INET6:
2434                 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0))
2435                         xprt_set_bound(xprt);
2436
2437                 INIT_DELAYED_WORK(&transport->connect_worker,
2438                                         xs_tcp_connect_worker6);
2439                 xs_format_peer_addresses(xprt, "tcp", RPCBIND_NETID_TCP6);
2440                 break;
2441         default:
2442                 kfree(xprt);
2443                 return ERR_PTR(-EAFNOSUPPORT);
2444         }
2445
2446         if (xprt_bound(xprt))
2447                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2448                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2449                                 xprt->address_strings[RPC_DISPLAY_PORT],
2450                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2451         else
2452                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2453                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2454                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2455
2456
2457         if (try_module_get(THIS_MODULE))
2458                 return xprt;
2459
2460         kfree(xprt->slot);
2461         kfree(xprt);
2462         return ERR_PTR(-EINVAL);
2463 }
2464
2465 /**
2466  * xs_setup_bc_tcp - Set up transport to use a TCP backchannel socket
2467  * @args: rpc transport creation arguments
2468  *
2469  */
2470 static struct rpc_xprt *xs_setup_bc_tcp(struct xprt_create *args)
2471 {
2472         struct sockaddr *addr = args->dstaddr;
2473         struct rpc_xprt *xprt;
2474         struct sock_xprt *transport;
2475         struct svc_sock *bc_sock;
2476
2477         if (!args->bc_xprt)
2478                 ERR_PTR(-EINVAL);
2479
2480         xprt = xs_setup_xprt(args, xprt_tcp_slot_table_entries);
2481         if (IS_ERR(xprt))
2482                 return xprt;
2483         transport = container_of(xprt, struct sock_xprt, xprt);
2484
2485         xprt->prot = IPPROTO_TCP;
2486         xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
2487         xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
2488         xprt->timeout = &xs_tcp_default_timeout;
2489
2490         /* backchannel */
2491         xprt_set_bound(xprt);
2492         xprt->bind_timeout = 0;
2493         xprt->connect_timeout = 0;
2494         xprt->reestablish_timeout = 0;
2495         xprt->idle_timeout = 0;
2496
2497         /*
2498          * The backchannel uses the same socket connection as the
2499          * forechannel
2500          */
2501         xprt->bc_xprt = args->bc_xprt;
2502         bc_sock = container_of(args->bc_xprt, struct svc_sock, sk_xprt);
2503         bc_sock->sk_bc_xprt = xprt;
2504         transport->sock = bc_sock->sk_sock;
2505         transport->inet = bc_sock->sk_sk;
2506
2507         xprt->ops = &bc_tcp_ops;
2508
2509         switch (addr->sa_family) {
2510         case AF_INET:
2511                 xs_format_peer_addresses(xprt, "tcp",
2512                                          RPCBIND_NETID_TCP);
2513                 break;
2514         case AF_INET6:
2515                 xs_format_peer_addresses(xprt, "tcp",
2516                                    RPCBIND_NETID_TCP6);
2517                 break;
2518         default:
2519                 kfree(xprt);
2520                 return ERR_PTR(-EAFNOSUPPORT);
2521         }
2522
2523         if (xprt_bound(xprt))
2524                 dprintk("RPC:       set up xprt to %s (port %s) via %s\n",
2525                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2526                                 xprt->address_strings[RPC_DISPLAY_PORT],
2527                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2528         else
2529                 dprintk("RPC:       set up xprt to %s (autobind) via %s\n",
2530                                 xprt->address_strings[RPC_DISPLAY_ADDR],
2531                                 xprt->address_strings[RPC_DISPLAY_PROTO]);
2532
2533         /*
2534          * Since we don't want connections for the backchannel, we set
2535          * the xprt status to connected
2536          */
2537         xprt_set_connected(xprt);
2538
2539
2540         if (try_module_get(THIS_MODULE))
2541                 return xprt;
2542         kfree(xprt->slot);
2543         kfree(xprt);
2544         return ERR_PTR(-EINVAL);
2545 }
2546
2547 static struct xprt_class        xs_udp_transport = {
2548         .list           = LIST_HEAD_INIT(xs_udp_transport.list),
2549         .name           = "udp",
2550         .owner          = THIS_MODULE,
2551         .ident          = XPRT_TRANSPORT_UDP,
2552         .setup          = xs_setup_udp,
2553 };
2554
2555 static struct xprt_class        xs_tcp_transport = {
2556         .list           = LIST_HEAD_INIT(xs_tcp_transport.list),
2557         .name           = "tcp",
2558         .owner          = THIS_MODULE,
2559         .ident          = XPRT_TRANSPORT_TCP,
2560         .setup          = xs_setup_tcp,
2561 };
2562
2563 static struct xprt_class        xs_bc_tcp_transport = {
2564         .list           = LIST_HEAD_INIT(xs_bc_tcp_transport.list),
2565         .name           = "tcp NFSv4.1 backchannel",
2566         .owner          = THIS_MODULE,
2567         .ident          = XPRT_TRANSPORT_BC_TCP,
2568         .setup          = xs_setup_bc_tcp,
2569 };
2570
2571 /**
2572  * init_socket_xprt - set up xprtsock's sysctls, register with RPC client
2573  *
2574  */
2575 int init_socket_xprt(void)
2576 {
2577 #ifdef RPC_DEBUG
2578         if (!sunrpc_table_header)
2579                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
2580 #endif
2581
2582         xprt_register_transport(&xs_udp_transport);
2583         xprt_register_transport(&xs_tcp_transport);
2584         xprt_register_transport(&xs_bc_tcp_transport);
2585
2586         return 0;
2587 }
2588
2589 /**
2590  * cleanup_socket_xprt - remove xprtsock's sysctls, unregister
2591  *
2592  */
2593 void cleanup_socket_xprt(void)
2594 {
2595 #ifdef RPC_DEBUG
2596         if (sunrpc_table_header) {
2597                 unregister_sysctl_table(sunrpc_table_header);
2598                 sunrpc_table_header = NULL;
2599         }
2600 #endif
2601
2602         xprt_unregister_transport(&xs_udp_transport);
2603         xprt_unregister_transport(&xs_tcp_transport);
2604         xprt_unregister_transport(&xs_bc_tcp_transport);
2605 }
2606
2607 static int param_set_uint_minmax(const char *val, struct kernel_param *kp,
2608                 unsigned int min, unsigned int max)
2609 {
2610         unsigned long num;
2611         int ret;
2612
2613         if (!val)
2614                 return -EINVAL;
2615         ret = strict_strtoul(val, 0, &num);
2616         if (ret == -EINVAL || num < min || num > max)
2617                 return -EINVAL;
2618         *((unsigned int *)kp->arg) = num;
2619         return 0;
2620 }
2621
2622 static int param_set_portnr(const char *val, struct kernel_param *kp)
2623 {
2624         return param_set_uint_minmax(val, kp,
2625                         RPC_MIN_RESVPORT,
2626                         RPC_MAX_RESVPORT);
2627 }
2628
2629 static int param_get_portnr(char *buffer, struct kernel_param *kp)
2630 {
2631         return param_get_uint(buffer, kp);
2632 }
2633 #define param_check_portnr(name, p) \
2634         __param_check(name, p, unsigned int);
2635
2636 module_param_named(min_resvport, xprt_min_resvport, portnr, 0644);
2637 module_param_named(max_resvport, xprt_max_resvport, portnr, 0644);
2638
2639 static int param_set_slot_table_size(const char *val, struct kernel_param *kp)
2640 {
2641         return param_set_uint_minmax(val, kp,
2642                         RPC_MIN_SLOT_TABLE,
2643                         RPC_MAX_SLOT_TABLE);
2644 }
2645
2646 static int param_get_slot_table_size(char *buffer, struct kernel_param *kp)
2647 {
2648         return param_get_uint(buffer, kp);
2649 }
2650 #define param_check_slot_table_size(name, p) \
2651         __param_check(name, p, unsigned int);
2652
2653 module_param_named(tcp_slot_table_entries, xprt_tcp_slot_table_entries,
2654                    slot_table_size, 0644);
2655 module_param_named(udp_slot_table_entries, xprt_udp_slot_table_entries,
2656                    slot_table_size, 0644);
2657